Search papers, labs, and topics across Lattice.
2
0
4
Robotic world-action models do not need pretrained internet video backbones to scale: training native planning and dynamics modules from scratch across 30,000 hours unlocks strong zero-shot manipulation and drives a 17.7-point gain on an embodiment comprising under 2% of the data.
Forget brittle multi-policy execution and manual resets: RoboClaw's "Entangled Action Pairs" let robots self-correct and learn continuously, slashing human intervention by over 50% while boosting task success.
